Operate independent instances.

Application routing

Use an external load balancer with health checks. Applications reconnect after an instance failure. Each live connection and its database session belong to one Accelerator instance.

Local state

Caches, queued writes and CDC progress stay with each instance. Caches refill from traffic after restart. Configure the engine’s required replication identity for every instance.

Configuration transfer

Use export and import to copy a consistent settings snapshot. Backups include secrets and scopes but exclude cached data, queued writes, logs and live CDC progress. Transfer them securely.

accel export /secure/backups/accelerator.sqlite3
accel --instance /srv/accelerator-b import /secure/backups/accelerator.sqlite3

Rolling maintenance

Route new connections away from an instance, allow active work to finish and verify queue drain before stopping it. Restart with validated configuration and confirm CDC, warm-up and application behavior before restoring traffic.